[ 
https://issues.apache.org/jira/browse/HADOOP-13276?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16674023#comment-16674023
 ] 

Steve Loughran commented on HADOOP-13276:
-----------------------------------------

Thinking some more here, retrying on noauth can make sense if the credential 
providers know that there's been a failure and they should try to re-retrieve 
secrets. In particular, IAM authenticator can do this.

passing the refresh() command all the way down should kick this off: but how to 
trigger the retry?>

# Invoker could take an interface to a callback to invoke on auth failures 
before the retry; S3AFS would trigger that refresh of its 
{{AWSCredentialProviderList }}. Similar to 
{{org.apache.hadoop.fs.s3a.Invoker.Retried}} but stored during construction 
# need to identify all auth failed exceptions as special & translate into a 
specific exception. 403, probably
# exception translation would need to invoke that iff there was an auth failure.

Taking a Retried callback in the constructor would be trivial; may want to add 
a new interface tho' which lets the callback actually indicate whether or not 
to retry. Why so? you could identify when an auth wasn't working as there'd 
just been a previous attempt. Realistically: only one failure on a callback is 
going to be recoverable. If, after a cred refresh nothing works, you may as 
well give up


> S3a operations keep retrying if the password is wrong
> -----------------------------------------------------
>
>                 Key: HADOOP-13276
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13276
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs/s3
>    Affects Versions: 2.8.4
>            Reporter: Steve Loughran
>            Assignee: Thomas Poepping
>            Priority: Minor
>
> If you do a {{hadoop fs}} command with the AWS account valid but the password 
> wrong, it takes a while to timeout, because of retries happening underneath.
> Eventually it gives up, but failing fast would be better.
> # maybe: check the password length and fail if it is not the right length (is 
> there a standard one? Or at least a range?)
> # consider a retry policy which fails faster on signature failures/403 
> responses



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to